A columnar database is a database having an extremely fast query speed. By dividing one table in a relational database into multiple columns and storing them separately, the columnar database allows an entire column of data to be read at one time during a database query. The columnar database has an advantage of reducing consumption of a database projection operation. During a query, it is not necessary to read data of an entire table, but only necessary to read a corresponding column, which dramatically reduces a read input/output (IO) amount. However, due to a storage manner of a columnar database, a write operation of the columnar database is slower than that of a traditional row-oriented database.
An existing columnar technology is to obtain columns in a relational database by means of division and store them separately. In this case, all columns need to be modified during row writing. However, because the columns are stored after division, addressing needs be performed multiple times during row writing on a hard disk, and write efficiency is equivalent to that of random write. A speed of the random write is far slower than that of sequential write, and therefore, write efficiency of the columnar database is extremely low.